One server goes down. Nobody can buy anything.

Understand it one step at a time

The short runs these in order in about 32 seconds. Here they are written out — pick any step to jump the short straight to it.

1. 1

   Millions of shoppers, one shop

   Logins, carts and checkouts — every one of them lands on the same server.

   all systems normal

2. 2

   Everything depends on that one box

   Traditional design: users, one server, one database. No spare, no second copy.

   single point of failure

3. 3

   Maintenance starts. The server goes offline

   A patch, a reboot, a failed power supply — it does not matter which. The box stops answering.

   server unreachable

4. 4

   Customers cannot log in

   The architecture diagram fails silently. Your customer sees a spinner, then an error, then leaves.

   sign-in failing

5. 5

   This is what the outage costs

   Lost customers, lost revenue, damaged reputation — the bill keeps growing while the box reboots.

   business impact

6. 6

   Azure splits a region into zones

   Every zone-enabled Azure region has three or more physically separate datacenters — own power, own cooling, own network.

   availability zones

7. 7

   Run the shop in all three

   One load balancer in front, an app instance in every zone, and the database replicated across them.

   zone-redundant deployment

8. 8

   Zones share data synchronously

   Under 2 ms apart, so a write can wait for all three. The commit is only acknowledged once every zone has it — zero data loss.

   synchronous replication · RPO 0

9. 9

   Zone 2 fails. The shop stays open

   The load balancer stops sending traffic there within seconds. Zones 1 and 3 absorb it — logins keep working.

   failover in progress

10. 10

    One server: 99.9%. Three zones: 99.99%

    Azure backs it with an SLA — 8.7 hours of downtime a year becomes 52 minutes, for the price of spreading out.

    zone redundant · SLA 99.99%